Research On Agricultural Disaster Area Forecasting Method Using PSO-KSVM For Middle Of Hunan Provence

China is one of the countries to have the worst natural disasters all over the world.There are a few characteristics about natural disasters,which often happen frequently and widespread,which also have many types and cause heavy losses.Through the analysis of the previous natural disasters in our country, we found that 70% of the natural disasters belong to meteorological disasters,because our country agricultural infrastructures are weak, the anti-disaster ability is not strong enough, the agricultural production highly depends on the climate. All these result in covering the entire disaster- stricken of annual crops more than 50 million hm2 and the economic loss of more than 200 billion yuan.Currently, global climate has changed abnormally and new changes have been made in agricultural disaster, so the prevention and mitigation of agricultural disaster remain a long way to go.Therefore, based on the characteristics of middle Hunan agricultural disaster, and the combination of the Predicting Model of Support Vector Machine, Kernel Function Theory, Particle Swarm Optimization, then to study and apply the predictive methods of disaster area.The research work includes several parts as fellow:(1)Collect the datas and use the analysis technology of the outlier datas, remove abnormal datas and make the standardization process.Because Loudi was founded a short time ago and it,s historical datas are limited,the datas in this paper are collected from 1975 to 2015 in "Hunan Rural Statistical Yearbook" and "Loudi City Agricultural Bureau of agricultural statistics annual report". By analyzing outlier we found that because of the special and anomal weather,heavy losses are caused during three years, namely 1998, 2008, 2012 and significant losses were specially deal with.We remove it as abnormal datas and finally makes normalized processing to the rest datas.(2) Put forward a kind of agricultural disaster area prediction method based on KSVM.Kernel function selection is mainly three types:Polynomial kernel function, Radial basis function and Sigmoid kernel function. Based on the past experience, we use Radial basis kernel function, and because of less parameter, if the parameter is appropriate, it will be more suitable for forecasting model of this article.So we choose Support vector machine(SVM) model,which using Radial basis function as the kernel function. Finally, compared with the BP neural network prediction model,we can find that when we use Radial basis function(RBF) as kernel function,Support vector machine(SVM) model will be more suitable for small sample environment.(3) Put forward a prediction method of agricultural disaster area based on PSO-KSVM.The parameter choosing according to the Radial basis function kernel of SVM is not the best strategy. By implementing Particle Swarm Optimization and optimizing the parameters, I built a prediction model of agricultural disaster area in middle of Hunan province based on PSO-KSVM. Thereby, I compare the result of the prediction model and BP network model on the same specimen, and I obtain the result that the prediction model achieve a better performance with higher precision.